|REPRINTED WITH PERMISSION FROM||
The Vacuum Tube & the Human Ear
By Keith Herron
Like many people, I dumped my tube gear back in the early Seventies in favor of the "new" solid-stage technology. Wow, the solid state stuff had such impressive specifications! Humongous power, zero point-oh-oh-ohhhhhh distortion figures, flat from DC to channel 5. What's more, it doesn't get hot, it won't wear out, requires no maintenance, and it is ready to go the instant you turn it on. I was so caught up in the technology, I might have missed something in the sound. I spent most of the Seventies and Eighties designing with solid state.
I was introduced to the vacuum-tube revival as a director of engineering for a musical instrument amplifier manufacturer in the late Eighties. A growing number of electric guitar and bass playing musicians were demanding "tube" sound and buying tube amplifiers. It was a growing trend, parallel in a sense to what was happening concurrently in High End audio. Along with producing tube guitar amps, heads, and combos, we produced a number of solid-state amps designed to mimic the response of tubes -- with mixed results. Clearly the tube revival had a lot more to do with the sound than it did with nostalgia. Assuming you know some of the basics about tubes, let's look a little deeper into the technology and into human hearing to explore the sonic effects of tubes.
Our sensitivity to frequency response
The advances in technology of the past several years provide us with the tools necessary to acquire a much better understanding of the phenomenon at work in the electronics of the reproduction of sound.
Precise measurement coupled with experienced listening can produce repeatable correlations between what is heard and what is measured. This is by no means a new concept. What has changed are the tools we have available for measurement and analysis. The results of such analysis provide surprising information about the sensitivity of our hearing to seemingly minute variances in that which we can measure. One area of analysis we have found most fascinating is the sensitivity of our ears to variation in frequency response.
During the development of the VTPH-1 phono stage, a number of correlations were made between measurement and audible perception. One process involved a highly accurate mathematical analysis of the RIAA response curve and development of procedures to resolve measurements down to one one/hundredth of a decibel (millibel). Data from these measurements were compared to listener evaluations of several units done in several listening systems. Switching filter network parts between units allowed us to reverse the listener responses as to most favored and least favored on a repeatable basis (the sound followed the parts). The only difference in these parts were variations in value due to the tolerance. Parts of identical value were substituted in order to make sure that no other phenomenon was involved.
We found that a 3-millibel variation in the frequency response curve can change a listener's perception of the unit. This could explain the audible improvements often associated with wide bandwidth in audio equipment. A loss of frequency response in terms of decibels at 100kHz will mean, in most cases, a drop in frequency response of several millibels at 10kHz - a perceptible change. Many of the colorations (sonic variations) heard in audio equipment will correlate with frequency response variations. It is somewhat surprising that one can consistently hear such small differences between electronic components when the listening is done with speakers that vary in response by several decibels.
Filters that alter frequency response take many forms in audio circuitry. The most basic in analogue designs involve resistors, capacitors, and inductors. The combination of two or more of these in a circuit will produce a filter. One of the most common problems an audio designer runs into is the fact that many of the resistors, capacitors, and inductors in a circuit are not the filter components purchased to be installed in the circuit. They end up being resistance in component leads and circuit traces, circuit output impedances, capacitances between conductors and component parts, and inductances associated with circuit conductors. Capacitances occur inside tubes, as well as solid-state amplifying devices. The grid-to-plate capacitance inside a triode is generally on the order of a few picoFarads for small signal preamplifier tubes. This capacitance will act as a high-frequency roll-off filter if there is any impedance (resistance) feeding the control grid. To make matters worse, the effective capacitance of the filter is the product of the grid-to-plate capacitance and the gain of the tube.
This product is known as the Miller effect. As the voltage on the control grid is either increased or decreased, the plate voltage makes large changes in the opposite direction by a factor equal to the voltage gain of the tube circuit. The effective grid-to-plate capacitance becomes a form of negative feedback and reduces response at high frequencies. A 12AX7 will have 1.7 picoFarads grid-to-plate capacitance and will act like 170 picoFarads if the voltage gain of the circuit is 100. This will produce a roll off of 3dB at 20kHz when a 47k-source impedance is used at the control grid. This circuit will be down 98 millibels at 10kHz. And that's far more than the 3 millibels frequency response variations we were able to detect in our listening tests.
It becomes clear, therefore, that replacing a tube with one that has a slightly different gain or grid-to-plate capacitance will alter the sound. Circuit designs can compensate for some of this by lowering the source impedance and other techniques.
Many triode circuits, as a result of the Miller effect, will not have very good high-frequency response where tubes with high grid-to-plate capacitance and high voltage gains are used.
I recently encountered a line stage preamplifier with a 470k/ohm volume control feeding into a 6SN7 twin triode that has a grid-to-plate capacitance of 4 picoFarads with a gain factor of 20. The effective capacitance (Miller effect) would be 80 picoFarads. The maximum source impedance to the grid of the tube from the 470k volume control will be at least 235k, when the volume control is 6 dB down from full output. This will result in a 3 dB down point for the circuit under these conditions at roughly 8.5kHz.
Is that rolled off or what? Results: a very polite and smooth high end, but no detail. As the volume control position is changed, the corner frequency of the roll off filter changes. This is true to some extent with most line-stage designs, tube or solid state. Try doing comparisons with two phono stages that have very different gain factors. If moving the volume control on the line stage changes the sound, what will be the results of the comparison? It is generally true that the line stage will have the widest frequency response as the volume control is turned down toward the minimum. The higher gain phono stage will either have better highs or be too bright, depending on the quality of the equipment involved.
A wide variety of power amplifier configurations have come and gone over the years and, yes, some that were thought to be gone have come back. Although tube amplifiers never completely left the scene, solid-state designs were dominant during the Seventies and Eighties. Demand for tube amplification equipment for High End audio has been on the rise in the Nineties. The most common tube configurations are push-pull (using power pentodes or triodes), single ended (using mainly triodes), and output transformerless (OTL) designs.
If it's lots of power you want, and you want to get it with tubes, you will most likely end up using an amplifier with power pentodes or beam power tubes in a push-pull configuration and a transformer at the output. This configuration uses at least one power tube for each half (positive and negative) of the sine wave. The output tubes for each half of the waveform are driven with signals from opposite sides of a circuit known as a phase splitter. This shifts the phase between the output tubes by 180 degrees. One side pushes while the other pulls (or takes a break, in the case of Class B operation). The output transformer windings correct the phase per half-cycle at the output.
The single-ended configuration generally uses the output tubes (one or several in parallel) in Class A operation. The advantage of this configuration is that the output tubes operate through the entire audio waveform and there will be no switching between tubes and no phase splitter needed for driving two halves of the output stage. This type of circuit is not very efficient and dissipates a significant amount of heat. A single-ended Class A triode output stage will usually generate more second-harmonic distortion than does a push-pull configuration that tends to cancel out even order distortion products.
To me, as an engineer, it seems inconceivable that a tube amplifier with a big old transformer at the output could sound better than a solid-state amplifier with no transformer.
The output transformerless (OTL) configuration dispenses with the output transformer, as its name implies. The trick here is to get enough current directly from tubes to drive speakers, and to do it without having DC at the output. The purpose of an output transformer is to transform the ratio of voltage to current at the plates of the output tubes (high voltage, low current) to a ratio that will deliver adequate power to the speakers (high current, low voltage). The transformer is an impedance matching device.
An output transformer designed to deliver reasonable power at low frequency requires a relatively large iron core. In order to deliver lower frequency response at a given power level, more iron is required in the magnetic circuit of the transformer to prevent saturation. This increases the size of the transformer. The larger the transformers, the longer the wire in each winding; more resistance is introduced by having longer wire. The additional wire tends to reduce high-frequency response. Transformer design is plagued with trade offs - many more than I mention here.
So how do you build a vacuum tube power amplifier without an output transformer? You need to use tubes with high current capability and a lot of them. If tubes with lower current capability are used, then more will be needed. Special circuit designs are needed in order to keep the output impedance low. A fair amount of feedback is often a requirement.
All of the configurations discussed here, unless provided with automatic self-biasing or cathode-bias resistors, will require the user to periodically adjust the output tube bias. Failure to do so may result in either increased distortion with possible loss of power, or excessive heat and shortened tube life. Configurations that require more than one output tube should be used with matched sets of tubes. Mismatched tubes can cause increased distortion and shortened tube life. Biasing is basically adjusting the idle current of the tube by setting a negative voltage (bias) on the control grid of the tube.
When triodes are used as output tubes, additional steps must be taken in order to achieve good high-frequency response. The high grid-to-plate capacitance of triode output tubes and the resulting Miller effect will require the control grid to be driven with sufficient current to overcome the effective capacitive reactance at the grid. This often requires pentode power tubes to be used in the driver stage.
Tube substitutions should be done only in institutions. In a recent discussion with my tube supplier, I asked, "What are the differences are between the Sovtek 12AX7WAs and the Sovtek 12AX7WBs?" He replied, "The WAs have lower gain but many people don't like them as well because they are brighter." This was not a surprising answer. An otherwise identical tube with slightly less gain would exhibit less Miller-effect capacitance and therefore a wider bandwidth. This would produce more highs, at least by a few millibels -- which, as we have demonstrated, people can hear.
The lesson here is simple: You cannot just plug different tube types into a given circuit and evaluate them. They have different electrical properties and therefore require different circuits to obtain the best performance! A Sovtek 12AX7WA, for example, with its lower gain, will have less Miller effect equivalent capacitance and therefore a wider bandwidth than the Sovtek 12AX7WB. It would be brighter in the wrong circuit. The difference in gain alone will change the frequency response. In the right circuit, the 12AX7WA just might produce the best sound of the two.
Different tube types (power tubes included) cannot be plugged into the same circuit and properly compared, even with the bias readjusted. The various tube types have different inter-electrode capacitances, plate resistances, output transformer matching requirements, transconductances, etc. One would very likely come to the wrong conclusions about which tubes could really give the best sound under the right operating conditions. Many people make such comparisons with a total lack of understanding of the nature of tubes.
Comparison of various tubes of the same type does have some merit. Understand, however, that various manufacturers of a particular tube type will not produce tubes with the exact same electrical parameters. The sound will usually be quite different owing to variations in electrical response, microphonics, and internal resonances excited by electrostatic forces in the tube. Harmonic content will also vary from brand to brand. Considering that much of the assembly involved in the manufacture of vacuum tubes is done by hand, it is not surprising that tubes with the same part number from a particular manufacturer and production run have audible variations.
A word on solid state
In the solid-state world, the field effect transistor (FET) and metal oxide field effect transistor (MOSFET) have properties most like that of a tube. They both convert voltage changes at the gate (input) to current changes at the drain and source (output). Like tubes, neither draws significant current from the input source at low frequency. Bipolar junction transistors (BJTs) are the most common solid-state devices used in preamplifiers and power amplifiers. The BJT amplifies current and in doing so requires some current from the input source to drive the device. Any non-linearity in the operation of a BJT will produce distortion at its source.
Many typical common emitter implementations of bipolar junction transistors will produce predominately odd-order distortion at a relatively low, but somewhat constant, percentage of the signal level. Odd-order harmonics are not as much of a natural phenomenon as even-order overtones in music. An unloaded triode, on the other hand, will produce predominantly even-order harmonics at levels proportional to the signal level. The odd-order harmonics produced start very small and increase with the square of the signal level. Even or odd, the lower the signal level with tubes, the lower the distortion. The bottom line is that a properly implemented triode will produce very low distortion at the small signal levels where much of the detail exists in recorded music. In many cases, at these levels it can be much lower than the oh so impressive, but constant distortion figures of many bipolar transistor designs.
Much of the solid-state audio gear that I have heard, particularly the stuff that employs BJTs in the signal path, tends to sound more aggressive and harsh though the midrange and high frequencies. The sound stage generally has less depth, width, and height than that of its vacuum-tube counterpart. I have encountered a few designs that are exceptions to this and that makes me wonder if the sonic differences I have heard have more to do with proper or improper implementation of the various amplifying devices than with any guaranteed certain results of their use.
The worst thing an audio engineer can do is to assume that he or she knows much about the nature of human hearing. After all, the engineer who designed our hearing knows a few things we never will.
Keith Herron is the owner and principal designer at Herron Audio and bona fide E.E. He was formerly director of R&D for Ampeg, Crate, and Audio Centron. He combines a strong background in electronics, math, physics, and music with a keen ear and open-minded analysis of audio equipment design. He describes himself as an audiophile, a lover of music, and a "former" musician.
To subscribe to the absolute sound, call 888-732-1625 (US), 973-627-5162 (outside US) or visit www.theabsolutesound.com. $42 for six issues in the US; $45 Canada, $75 outside North America.
Reprinted by permission from Absolute Multimedia, Inc. All rights reserved. Any unauthorized duplication of this article is strictly prohibited.
Copyright © 1996 - 2021 Herron Audio. All rights reserved.
Herron Audio, div. of Herron Engineering, Inc.
12685 Dorsett Road #138, Maryland Heights, MO 63043 U.S.A.